Dancing the Radar Sensitivity Limbo: How Low Can We Go?

One way to improve radar data assimilation is to develop new increase the coverage of radar data, primarily by trying to detect weaker echoes. To that end, we implemented and tested the skill of more advanced signal processing algorithms in an attempt to enhance sensitivity, such as poly-pulsepair techniques and adaptive coherent-averaging techniques targeted at enhancing the returns from specific velocity ranges. What was found was that more advanced processing algorithms do not always improve estimation accuracy, but instead have a more bimodal behaviour: They either succeed to make a better estimate, or they fail miserably and make things much worse. The trick is hence to be able to recognize which occurs, and it has put to the forefront the importance of properly determining good versus bad estimates. In that process, we have also realized that current thresholding algorithms based on power or SQIs are very wasteful, and in our attempt to limit false alarms to very few occurrences, we throw out a lot of valid measurements.

In an attempt to rescue more of that thresholded-out data, we computed velocities over large areas as constraints for smaller-scale estimates. The basis for that approach is that weak signals generally occur in benign weather where winds are not expected to be very variable in space and that such an average could be used as a reference value that smaller-scale estimates should not deviate too much from to be considered acceptable. But as sensitivity increases, so does our ability to detect both weather and clutter. We hence experimented computing spectra that use both horizontally- and vertically-polarized returns simultaneously in a way that spectral components dominated by weather, where H & V signals are nicely correlated, add themselves coherently, while they do not for noise, and do so partially for clutter. This behavior helps us 1) eliminate clutter- and noise-contaminated spectral elements and 2) better compute the reference velocity that may help the thresolding process. Finally, we are also investigating less wasteful ways of doing the actual thresholding.